C2PA Content Credentials: The New Metadata Arriving Inside Your Photos
Published 4 September 2026 ยท 7 min read
EXIF is old technology. It was designed in the 1990s for cameras that had no idea what the internet would do with location tags and serial numbers, and we have covered its privacy problems at length, from lingering GPS data to camera serial numbers that fingerprint every file. A newer standard is now arriving in mainstream devices, and it works differently enough that it deserves its own explanation. It is called C2PA, its consumer-facing name is Content Credentials, and if you have bought a recent phone or camera there is a fair chance it is already embedded in some of your images.
What C2PA actually is
C2PA stands for Coalition for Content Provenance and Authenticity, a joint effort backed by Adobe, Microsoft, Intel, BBC, Sony and others, with the stated goal of fighting misinformation by recording where a piece of media came from and how it was edited. Technically it is a signed manifest: a block of data embedded in the image file (in JPEG, PNG and increasingly in a sidecar JUMBF structure) containing assertions such as what device captured the image, what software edited it, which generative AI model produced or altered it, and a cryptographic signature chaining those assertions together.
That last part is the key difference from EXIF. EXIF fields are free text that anyone can write, which is why fake metadata is trivial. C2PA assertions are signed with private keys held by the manufacturer or software vendor, so in principle a forged manifest fails signature verification. In practice the ecosystem has cracks, private signing keys have leaked from major vendors more than once, but the design intent is verifiable provenance rather than self-declared provenance.
Where you will meet it
OpenAI, Google and Adobe attach C2PA manifests to AI-generated images. Sony and Leica ship it in camera bodies. Recent Android releases from major manufacturers embed it by default in the camera pipeline. Photo editing software increasingly appends an edit-history assertion every time you save. Social platforms strip some or all of it on upload, inconsistently, which is its own problem.
For an ordinary user the practical consequence is this: your photos can now carry a signed, timestamped record of the device that took them and every tool that touched them since. For provenance verification that is powerful. For privacy it is a second metadata channel running alongside EXIF, and one that most stripping tools do not yet look at.
Why this matters for privacy
The uncomfortable overlap is device identity. A C2PA capture assertion typically includes the make, model and sometimes a device identifier, signed by the manufacturer. That is the same class of information as the EXIF BodySerialNumber we described in our piece on serial number metadata, except harder to remove cleanly, because a removed manifest is itself detectable: the absence of a signature where one would be expected is a signal.
Edit history cuts the other way and is worth thinking about before you share. A screenshot you crop before sending may carry an assertion describing the crop, the tool, and the timestamps. People who crop out sensitive parts of an image, a street sign, a face, a document corner, generally assume the crop is the whole story. With Content Credentials the manifest can say that a crop happened even when the cropped pixels are gone.
It also interacts badly with the habits most people already have. As we showed when looking at what messaging apps do with metadata, sending an image as a file preserves far more than sending it as an inline image. C2PA manifests survive that same path, and they survive re-encoding better than raw EXIF in some container formats because they can live in a dedicated box that generic re-encoders copy forward.
The double bind of stripping
Here is the honest tension. If you strip C2PA entirely, your image looks like every other metadata-free image on the internet, which is normal and unremarkable. If a platform or recipient specifically expects signed provenance and finds none, they cannot tell the difference between a privacy-conscious photographer and someone hiding an AI generation. Nobody is currently penalised at scale for missing manifests, so for privacy purposes stripping remains the sane default. The double bind matters mainly for journalists and creators who want the credentials for verification while also wanting location privacy, and the standard does allow partial redaction: capture assertions can be retained while GPS and device identifiers are removed. Few consumer tools expose that control yet.
What to actually do
- Check files before sharing, including for C2PA, not just EXIF. You can drag any image onto the reader on our homepage and inspect what is really inside, locally, with nothing uploaded.
- Prefer platform inline sending over file sending when you do not control the receiving end; both metadata channels survive file transfers more often than they survive platform processing.
- Assume edit history travels with the file. Sensitive crops deserve a round trip through a tool that re-encodes cleanly, and verification after.
- If you publish photography professionally, decide your provenance stance deliberately: full credentials, partial redaction, or clean files. Doing it file by file guarantees inconsistency.
Content Credentials were built to answer a real question, where did this image come from. That question has a legitimate owner in the verifier. But the same data answers another question, where did this person come from, and that one belongs to you. Treat the manifest as you learned to treat EXIF: read it before it leaves.